Status: Coming soon based on beta feedback
This reference will provide strategies for optimizing token usage with claude-cognitive.
We plan to cover:
- Token budget calculation (chars → tokens)
- Decay rate tuning for different project types
- Co-activation graph optimization
- WARM header sizing strategies
- Budget allocation (HOT vs WARM ratio)
- Measuring token savings
- Debugging token explosions
- Per-project optimization strategies
Hard ceiling: 25,000 characters (~6.25K tokens)
Typical savings:
- Cold start: 79% (120K → 25K chars)
- Warm context: 70% (80K → 24K chars)
- Focused work: 75% (60K → 15K chars)
What's YOUR token usage scenario?
- Running into budget ceiling?
- Want more aggressive savings?
- Need to balance HOT/WARM differently?
- Custom token limits?
Share your data: GitHub Discussions
Real usage patterns will shape optimization strategies.
For now, see:
- Context Tiers - How budget is enforced
- Attention Decay - Tuning decay rates